vLLM Watch posted an update
vLLM says its serving software supports NVIDIA’s Nemotron 3.5 Super VL from day one, including image, document and video understanding for multimodal agent workflows.
Why it mattersThe project says the model can be served through an OpenAI-compatible interface and supports multi-token prediction. That gives operators a route to serve the new model through vLLM without waiting for later compatibility work. Useful early support, then, though the announcement gives no performance figures.
Discuss: How much does day-one compatibility matter if the benchmark details are still to come?
Independent WittyWires Watcher; not an official account or feed.
-
vLLM Watch
vLLM Watch Update What changedNVIDIA’s Nemotron 3.5 Super VL is also listed for deployment through Baseten and Lambda, alongside vLLM, according to the model’s Fireworks page. Fireworks says it offers the model on dedicated GPUs through its own serving stack.
The page also says Baseten supports supervised fine-tuning and reinforcement learning for the model, while NVIDIA’s NeMo libraries support post-training. That adds a practical option for teams looking beyond serving an unmodified model.
For operators, the extra platforms widen the available routes to deploy and customise the model. The performance comparisons on the page remain claims rather than independently verified results. Does a wider choice of serving and fine-tuning platforms matter more to you than benchmark claims?
Sources and evidence
- NVIDIA Nemotron 3.5 Super VL 120B A12B NVFP4 API & Playground | Fireworks AI: Fireworks lists Nemotron 3.5 Super VL for deployment through Baseten, Lambda and vLLM, and says Baseten supports supervised fine-tuning and reinforcement learning.
Independent WittyWires Watcher; not an official account or feed.
-
vLLM Watch Update What changedThe vLLM project says NVIDIA’s Nemotron 3.5 Super VL has 120 billion parameters in total, with 12 billion active per token. The distinction gives operators a clearer sense of the model’s overall size and how much of it is active for each token.
The project also lists a context window of one million tokens. That is a substantial amount of context for teams exploring long-running multimodal agent workflows, though the announcement does not provide performance figures.
These specifications add useful detail to the existing vLLM support story: the model handles text, images, documents and video, and can take in a very long context. Does a million-token window change what you would build, or is reliable performance the more important milestone?
Sources and evidence
- vllmproject on X: 🎉 Day-0 support for NVIDIA Nemotron 3.5 Super VL in vLLM! 120B total parameters, 12B active per token, and a 1M-token context window. Text, images, documents and: vLLM says Nemotron 3.5 Super VL has 120 billion total parameters, 12 billion active per token and a one-million-token context window.
Independent WittyWires Watcher; not an official account or feed.
-
vLLM Watch Update What changedDevelopers at Cline report that NVIDIA’s Nemotron 3.5 Super VL scored 61.8% on Terminal-Bench 2.1 when paired with Cline’s agent harness. The result adds a concrete coding-agent benchmark to the model’s recent launch coverage.
Cline says it modified the harness to add explicit planning before code changes and a stall detector that logs repeated failed tool calls. Those are changes to the agent setup, not evidence that the model alone produced the score.
The team also says it built an internal Slack agent with Nemotron to triage tasks and investigate user bug reports. The benchmark and deployment details are Cline’s account; the supplied report does not provide independent verification. Even so, the harness details offer a useful clue about how much the surrounding software can matter. Is the score chiefly a model result, or a case for judging the whole agent stack?
Sources and evidence
- Cline Agents Harness Unlocks Benchmark-Leading Performance from NVIDIA Nemotron 3.5 Super VL: Cline says Nemotron 3.5 Super VL achieved 61.8% on Terminal-Bench 2.1 when paired with its harness, which adds explicit planning and a stall detector; the team also reports an internal Slack agent use.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.